Arctic cyclones are a fundamental dynamical driver of transient atmosphere heat and moisture transport between the Arctic and lower latitudes. They also modulate turbulent heat, moisture, and momentum fluxes across the atmosphere-ocean/sea ice/land interface. All of these are crucial elements controlling Arctic and global energy and water cycles. It is therefore important for climate models to realistically simulate cyclone activity. In this study, we analyzed Arctic cyclone activity in the CMIP6 historical simulations and compared the results with those obtained from the reanalysis datasets. The data we used are the 6 hourly sea level pressures (SLPs) from 31 earth system models. We employed a recently improved cyclone identification and tracking algorithm to detect cyclones. The results indicate that all models generally capture the major characteristics of the cyclone tracks over the North Atlantic and North Pacific Oceans as well as cyclone propagation into the Arctic. However, cyclone counts differ across the models, depending on model resolutions. We also examined long-term changes in Arctic cyclones using an integrative cyclone activity index (CAI). Although most of the models demonstrate positive trends over the historical period 1950-2014, they generally underestimate the rate of increase and show a large spread compared with the mean trend from three reanalysis datasets (ERA5, NCEP-NCAR, and JRA-55). In particular, one model largely overestimates the CAI linear trend, while another model shows a large negative trend. To explore the reasons for the model spread and biases, we further analyzed the driving mechanisms in each model and compared them with the reanalysis data.